How we sampled
Two public surfaces were read in the same window. Group A came from publicly visible sheet pages, which yielded 31 wrapped ACBuy addresses. Group B came from a maintained directory list and yielded 62 source addresses. Anything that failed to complete during collection was left out of the table rather than guessed at.
Group A was requested twice in effect: first as the wrapper, then as the source address rebuilt from the marker found inside it. Group B was requested once, at its own marketplace address. Both groups were contacted under one setting, with at most two requests open at a time and a gap between calls. The checking method page spells out that setting in full.
The window was not unlimited. One host stopped accepting connections partway through, and 113 directory addresses never completed. Those 113 are excluded, not written down as unknown. That choice matters more than it looks: a table that folds unfinished work into a failure column reads far worse than the evidence supports.
What the sample held
Every record in the period traces to Weidian: 93 of 93. Taobao, 1688 and unidentified sources each sit at zero. That is a property of the sample, not a claim about the market. The wrapped addresses we could collect all carried the same source marker, so the sample can only describe what that marker points at. Other markers exist in the wild; this site has no measured example of any of them.
The 62 directory records came from a list that removes entries it can no longer reach, so it describes what a maintained list looks like. The 31 sheet records came from pages nobody appears to have cleaned, so they describe what an unmaintained sheet looks like. Read side by side, the two groups say different things about maintenance and the same thing about the listings underneath.
What we counted
The status column reads 93 resolved, with moved, gone, shielded and unrecorded all at zero. A resolved status means the rebuilt source address answered when it was requested. It does not mean the item is in stock, that the seller is genuine, or that the page will answer tomorrow. A status records a response at a moment, nothing more.
The number worth sitting with is 31. Thirty-one wrappers, thirty-one failures at the wrapper, thirty-one normal responses from the addresses those wrappers encode. The break sits in the layer in front of the listing, not in the listing. Group B shows the other side of the same coin: 62 addresses, no cleaning involved, every one of them answering.
The field note on the "thousands of links" number takes that headline claim apart and shows how far the total moves depending on what you agree to count. Rows, unique items and working addresses are three different numbers, and a sheet page rarely says which one it is quoting.
What this cannot tell you
Counts describe 93 specific addresses. They say nothing about listings outside the sample, about wrappers whose source marker this site has never observed, or about how many links exist across all public sheets. Nobody has that last number, and a sheet page with thousands of rows has not told you how many of them still work.
Two limits are worth repeating. The 113 unfinished directory addresses are excluded, so the table is not a census of anything. And a resolved reading is a response, not a promise. Recount your own sheet rather than trusting any headline total, including the one here.